Stable carbon isotope analysis on fossil Cedrus pollen shows summer aridification in Morocco during the last 5000 years

Quantitative climate reconstructions from pollen typically rely on empirical relationships between pollen abundances or assemblages and climate, such as the modern analogue technique. However, these techniques may be problematic when applied to fossil sequences, as they cannot separate anthropogenic from climatic influence on pollen assemblages. Here, we reconstruct Mid‐ to Late Holocene summer aridity in the Middle Atlas, Morocco, using stable carbon isotope analysis of isolated fossil Cedrus pollen. This approach is based on well‐documented plant physiological responses to moisture stress and is therefore independent of vegetation composition. We find that there has been a general long‐term trend of increasing summer aridity in the region during the last 5000 years to the present day. The gradual decline of Cedrus atlantica forest in the Late Holocene follows this aridity trend. Additionally, we show how isolating a specific pollen type for carbon isotope analysis yields a robust climate signal, versus using pollen concentrates or bulk sediment. Our findings indicate that climate has become drier in the region and confirms the Mid‐ to Late Holocene aridification trend observed more widely in the western Mediterranean, using a novel proxy for this region with good potential for wider application in other environments.


Introduction
The Middle Atlas region of Morocco lies at the convergence of Saharan and Atlantic air masses, yielding a semi-arid to subhumid climate (Born et al., 2008). Fossil pollen assemblages from sediment cores taken across the Middle Atlas have provided insights into the region's past environment and climate (e.g. Lamb et al., 1989;Cheddadi et al., 1998Cheddadi et al., , 2015Cheddadi et al., , 2019Rhoujjati et al., 2010;Nour El Bait et al., 2014;Tabel et al., 2016;Campbell et al., 2017). A common pattern in these records is the virtual absence of Cedrus pollen in the Early Holocene, with the species generally becoming more prevalent from the Mid-Holocene.
The Middle Atlas is the largest core area of the North African Cedrus atlantica range despite deforestation during the recent millennia (Benabid and Fennane, 1994) and more recent dieback (Rhanem, 2011). The Mid-Holocene colonization of the Middle Atlas has been interpreted primarily as a signal of increased moisture availability (e.g. Cheddadi et al., 1998;Campbell et al., 2017;Zielhofer et al., 2017aZielhofer et al., , 2019. However, other factors such as winter temperature (Touchan et al., 2017), or ecological and edaphic conditions could have also influenced Cedrus expansion. Indeed, C. atlantica is a drought-tolerant species (but is sensitive to extreme drought; Aussenac, 1984;Rhanem, 2011) and is found growing in semi-arid mountain areas across North Africa (Fig. 1A), some of which have relatively low annual rainfall. Therefore, open questions remain about the role of climate in the establishment, persistence and decline of C. atlantica which may be difficult to answer based on pollen abundance data alone.
Quantitative climate reconstructions based on pollen assemblages typically rely on analogue matching (e.g. Modern Analogue Technique, MAT) or transfer functions to derive climatic inferences. However, these require a training set of modern assemblages and well-constrained pollen-climate relationships. Where the composition of either the fossil or the modern assemblages is strongly influenced by non-climatic factors including ecological succession, fire and anthropogenic activity (tree felling, farming, grazing) the approach can be problematic (Wahl, 2004;Ohlwein and Wahl, 2012). Accordingly, the increased impact of anthropogenic activity evident in Morocco during the Late Holocene (Zapata et al., 2013;Cheddadi et al., 2015;Campbell et al., 2017) which had widespread influence on vegetation composition, could potentially make pollen-based climate reconstructions using MAT unreliable, and other complementary approaches should be considered.
Climate reconstructions based on C. atlantica tree-rings should be less susceptible to anthropogenic influence. These show a multi-centennial scale pattern for the last~900 years with drier winters driven by positive winter NAO during the Medieval Climate Anomaly (MCA) (Trouet et al., 2009), wetter conditions during the Little Ice Age (LIA), and very recent dry conditions, with considerable interannual variability (Esper et al., 2007;Touchan et al., 2011). However, C. atlantica tree-ring-based studies cannot match the long-term perspectives offered by pollen studies, which could extend to at least the last 30 000 years in the Middle Atlas (e.g. Rhoujjati et al., 2010).
Here, we present the first stable carbon isotope analysis of isolated fossil Cedrus pollen from two sites in the Middle Atlas spanning the Mid-to Late Holocene. We demonstrate a methodology for the concentration and isolation of fossil pollen without the use of carbon-bearing chemicals. We also analyse pollen concentrates (organic residues rich in pollen) and bulk sediment samples from the same sample depths to evaluate the need for pollen isolation. C. atlantica is a climatically sensitive montane conifer endemic to North Africa of great biogeographical interest. Cedrus pollen are large (59.1 ± 4.0 µm) and easily identifiable, facilitating their study for isotope analysis (Bell et al., 2018). Furthermore, as an autumn pollinating species, pollen develops over the summer, making it possible to infer a summer climate signal (Bell et al., 2017).
There are well-established relationships between the stable carbon isotope composition of plant material and plant physiology (Farquhar et al., 1989). Photosynthesis allows plants to fix carbon (C); both 12 C and 13 C from CO 2 in the atmosphere through leaf stomata. Plants also regulate the rate of transpiration by changing the size of the stomata aperture to reduce water loss, allowing plants to adjust to different environmental conditions (Farquhar and Sharkey, 1982). In wetter conditions where stomata are fully open, 12 CO 2 and 13 CO 2 move freely between the atmosphere and leaf intercellular spaces. C fixation favours the lighter isotope ( 12 C) and intercellular 12 CO 2 is readily replenished as it is fixed; hence discrimination against the heavier isotope is greater. In drier conditions, stomata close to reduce transpiration, restricting the free movement of CO 2 . Consequently, C fixation is limited to CO 2 remaining in the leaf intercellular spaces, thereby increasing fixation of (or reducing discrimination against) the heavier isotope ( 13 C). Therefore, the ratio of 12 C to 13 C (δ 13 C) in plant tissues is dependent on environmental moisture availability at the time of C fixation. It is this relationship that allows a moisture availability signal to be reconstructed from δ 13 C of plant material (Farquhar et al., 1989;Dawson et al., 2002;Diefendorf et al., 2010).
As pollen δ 13 C is highly correlated with parent leaf material δ 13 C (Jahren, 2004), it can also be linked to environmental moisture availability. Early studies established that pollen δ 13 C reflects C3/C4 photosynthetic pathways (Amundson et al., 1997;Nelson et al., 2006;Descolas-Gros and Schölzel, 2007). Studies have also demonstrated a moisture signal in pollen δ 13 C of C3 plants (e.g. Loader and Hemming, 2004;Nelson, 2012;Griener et al., 2013). For C. atlantica, we previously demonstrated a strong moisture availability signal in modern pollen, and a strong relationship between pollen and parent leaf material δ 13 C (Bell et al., 2017). While most studies have focused on modern pollen, Griener et al. (2013) also studied fossil pollen to reconstruct moisture availability. However, applications of stable carbon isotope analysis on fossil pollen remain rare.
Several studies have detected a Late Holocene aridification trend from ecological or hydrological records in the western Mediterranean (Jalut et al., 2009;Carrión et al., 2010;Jiménez-Moreno et al., 2015). changes and stimulates the debate regarding the interaction of climatic and human impacts. Our summer aridity reconstruction allows us to evaluate the long-term aridity trends in the Middle Atlas from the Mid-Holocene to the present day, offering new climate insights based on pollen geochemistry.

Sample collection and preparation
Samples were obtained from sediment cores from two sites in the Middle Atlas, Morocco (Fig. 1B), including a deep perennial lake, Lake Sidi Ali (33.07°N, 5.00°E, 2080 m a.s.l.) Zielhofer et al., 2017a), and a shallow spring-fed wetland at Col du Zad (33.03°N, 5.07°E, 2148 m a.s.l.) (Campbell, 2017). The two sites are located within 5 km of each other and experience the same climate. The main wooded areas (including Cedrus atlantica) are located on the margins of Lake Sidi Ali (Fig. 1C), and a few kilometres from the Col du Zad core site. We hypothesize that while pollen assemblages can be influenced by local variability in the environmental setting, the isotope composition should be unaffected. The use of two sites allows us to test this and evaluate whether pollen isotope results from different sites can be integrated. Sample depths were chosen to cover the Mid-to Late Holocene period when Cedrus forests were well established, with one sample from the Early Holocene where a first interval of Cedrus expansion occurs. To obtain sufficient pollen for analysis, each sample integrated up to 4 cm of material for Lake Sidi Ali and up to 2 cm of material for Col du Zad, representing~20 and~30 years, respectively (Campbell, 2017;Fletcher et al., 2017).

Fossil pollen isolation
As traditional pollen preparation techniques (notably acetolysis) can alter the geochemical signal due to carbon-bearing chemicals (Loader and Hemming, 2004;Descolas-Gros and Schölzel, 2007;Nelson, 2012), a new technique based on micro-sieving (Heusser and Stock, 1984;Brown et al., 1989), sodium pyrophosphate washes (Bates et al., 1978) and dense media separation (Nakagawa et al., 1998;Campbell et al., 2016;Fletcher et al., 2017) was used to isolate pollen. The first stage involves concentration of the target pollen grains (summarized in Fig. 2 and described below), followed by the isolation (pollen picking) stage.
Up to 10 g of sediment was placed in a 50-mL centrifuge tube and disaggregated in 30 mL of 0.01 M sodium pyrophosphate (Na 4 P 2 O 4 ). Sediment was then sieved at 90 μm to remove coarse particles. The sub 90 μm material was washed by mixing with Na 4 P 2 O 4 and centrifuging for 3 min at 2200 r.p.m. and discarding the supernatant containing fine clays. This was repeated until the supernatant was clear, typically eight or more times. Samples were washed twice with deionized water (dH 2 O) to remove the Na 4 P 2 O 4 .
Samples were subsequently treated with 20 mL 10% HCl in a hot water bath for 15 min, or until effervescence stopped, to remove carbonates. Two density separations followed, the first using a sodium polytungstate (SPT) solution at 1.9 g cm −3 and a second separation using an SPT solution at 1.6 g cm −3 . The first separation aims to separate organic ( < 1.9 g cm −3 ) and mineral matter, while the second aims to concentrate pollen material in the lighter fraction (1.6 g cm −3 ). Each treatment involved centrifuging for 20 min at 1800 r.p.m. After the second density separation, the supernatant containing the light material was split into two clean 50-mL centrifuge tubes and topped with dH 2 O. This was centrifuged for 3 min at 4500 r.p.m., the two pellets were recovered and recombined into a single tube. Splitting into two tubes ensures that the density of the solution is sufficiently low as to retain all pollen in the pellets. This was washed thoroughly with dH 2 O to remove any remaining traces of SPT. The remaining sample was sieved at 32 μm to remove small pollen and other particles to concentrate the larger Cedrus pollen grains.
Pollen isolation was carried out on the concentrated pollen samples under a light microscope at 100 × optical zoom using a mouth pipette as described by Mensing and Southon (1999). dH 2 O and then transferred to a well slide using a micropipette (Fig. 3). Cedrus pollen grains were picked using the mouth pipette and transferred to a second well to isolate the grains from non-pollen material. Sometimes, other non-pollen material was transferred with the isolated grains into the second well. A final pick from the second well to be transferred into a small glass vial was used to remove this non-pollen material. Additional dilution with dH 2 O was applied as necessary for the final pick.
To ensure the pollen samples contained enough carbon for accredited δ 13 C results, a target number of 4000 grains was picked for each sample. Pollen picking time for each sample averaged 16 h, ranging from 12 to 21 h, with time taken largely dependent on the quantity of Cedrus pollen grains in the sample.

Pollen concentrates and bulk sediment
Preparation of pollen concentrates followed the method described above, except Cedrus pollen grains were not isolated from these samples. Bulk sediment was prepared by leaving in 30 mL 10% HCl for 24 h to remove carbonates, then washed twice with dH 2 O.

Stable isotope analysis
Stable carbon isotope analysis was carried out by the NERC Life Sciences Mass Spectrometry Facility at CEH Lancaster. Samples were dried at 105°C for 1 h and cooled in a desiccator, then weighed into small tin capsules for measurement. The isolated Cedrus pollen samples weighed between 0.040 and 0.075 mg. Capsules were combusted using an automated Eurovector elemental analyser coupled to an Isoprime Isotope Ratio Mass-Spectrometer (Elementar UK Ltd). An in-house working standard RefFLO (plain flour) was also analysed, resulting in a maximum analytical precision of 0.13‰ for isolated pollen, and 0.14‰ for pollen concentrates and bulk sediment. Data were normalized using in-house standards RefICE (cane sucrose) and RefGTAM (glutamic acid), calibrated annually against international standards (Sucrose-ANU -NIST 8542, and Glutamic acid -NIST 8573) and checked quarterly alongside the Wageningen proficiency testing samples within the International Plant Exchange (IPE) and International Soil Exchange (ISE) schemes. Three samples of each in-house standard are measured at the start and end of each sample run, used to construct a normalization calibration curve.

Sample size and preparation time
Theoretically, it is possible to analyse a sample containing a minimum of 5 µg carbon, although real world analysis often requires larger samples (Boutton, 1991). A series of tests using fresh C. atlantica pollen (~60% carbon) were run to determine the minimum quantity required for analysis. A theoretical minimum for the elemental analysis isotope ratio mass spectrometry (EA-IRMS) setup at CEH is 15 µg carbon. However, it was found that this minimum resulted in shifts to the known δ 13 C value of the reference material (sucrose) by 0.6‰, and the known δ 13 C of the pollen sample by 0.2‰, while the in-house standard remained unchanged. At 20 µg carbon, results were more consistent with expected δ 13 C values, but 30 µg carbon was found to be the 'optimal minimum'. As C. atlantica sporopollenin has a lower carbon content (~50% carbon) (Bell et al., 2017), a target minimum weight of 60 µg of fossil pollen grains (or~4000 grains) was required for analysis. The number of grains required for other pollen types and/or other laboratories is likely to vary.
This large number of pollen grains analysed has the benefit of providing a robust result that is representative of average environmental and climate conditions for the region. It may also be important to 'over-sample' due to potential losses of pollen grains between the preparation stages, e.g. transferring pollen grains from the glass vials to tin capsules for combustion inevitably results in small losses.

Data processing
All data and statistical analyses were carried out using R (R Core Team, 2016). For pure Cedrus pollen (modern and isolated fossil), δ 13 C values were converted to discrimination (Δ 13 C) using the formula: where δ A is the δ 13 C of atmospheric CO 2 , and δ B is the δ 13 C of the sample (Farquhar et al., 1989); this conversion corrects for changes in the δ 13 C of CO 2 over time (Schubert and Jahren, 2012). A second correction was made to the fossil pollen Δ 13 C values to additionally account for the impact of changes in the concentration of atmospheric CO 2 (pCO 2 ppmv), as detailed in Schubert and Jahren (2015;. Atmospheric CO 2 concentrations have varied from~260 ppmv in the Mid-Holocene to the present day~400 ppmv. These corrections are therefore important to allow robust comparison of fossil and modern isotopic values. Historical CO 2 δ 13 C and pCO 2 ppmv values were obtained from ice core records (Francey et al., 1999;Bauska et al., 2015;Eggleston et al., 2016). Full isotope data and corrections applied are available in the Supporting Information.

Aridity reconstruction
Aridity reconstructions were carried out using ordinary least squares regression, based on the relationship first established between the stable carbon isotope composition of modern C. atlantica sporopollenin and summer aridity in Bell et al. (2017). At the time of the original study, normalization of δ 13 C results against the Wageningen proficiency testing schemes was not used by CEH. Therefore, to ensure compatibility between the modern and fossil isotope results, six modern C. atlantica pollen samples were re-analysed at CEH. We used the new δ 13 C results (which differ by approximately −0.45‰) to apply a correction to the original data using the formula: δ 13 C corr = 1.0287*δ 13 C + 0.2724. We then re-ran the regression analysis to establish the relationship with aridity (Table 1).
Aridity data from the self-calibrating Palmer Drought Severity Index (scPDSI) were extracted from two gridded datasets: CRU v3.24 (Osborn et al., 2017) and Dai (2011). Due to differences in the calculation of these datasets, the severity of reported aridity can vary (Van Der Schrier et al., 2013). Although there is a stronger relationship with our isotope results and aridity data from Dai (2011), the CRU dataset is considered a more useful indicator of aridity in the region because it also accounts for snowmelt. Both sets of results are presented here for comparison with other published records. Due to the close proximity of the modern pollen sample sites (Bell et al., 2017) in Morocco, and the low resolution of the gridded aridity data: 0.5 × 0.5 and 2.5 × 2.5°for CRU and Dai (2011), respectively, the scPDSI data for Morocco were interpolated using bilinear interpolation.

Uncorrected δ 13 C results
Stable carbon isotope analysis shows δ 13 C has become less negative in isolated fossil Cedrus pollen during the last~5000 years, from a peak of −24.5‰ at~5100 cal a BP to −23.7‰ at 1150 cal a BP (Fig. 4A). At~8200 cal a BP, δ 13 C was −24.2‰; more negative than the most recent results but less negative than at 5100 cal a BP. The δ 13 C values from the two different sites are closely aligned.
Pollen concentrates show site-specific differences between the δ 13 C values, with Col du Zad having more negative δ 13 C values compared to Lake Sidi Ali, with peak low values at −27.7‰ at~5100 cal a BP (Fig. 4B). Lake Sidi Ali pollen concentrates yield similar δ 13 C values to isolated fossil Cedrus pollen, while Col du Zad values differ by around −2.6‰.
Bulk sediment also shows site-specific differences, with all Col du Zad δ 13 C values around −28‰, while Lake Sidi Ali δ 13 C values are between −22‰ and −25‰ (Fig. 4C). The Lake Sidi Ali bulk sediment δ 13 C values are also similar to isolated fossil Cedrus pollen. There is no overall trend in either the pollen concentrates or the bulk sediment results.

Cedrus pollen corrected Δ 13 C
In isolated fossil Cedrus pollen, discrimination has decreased during the last~5000 years (Table 2; Fig. 5C). Discrimination peaked at Δ 13 C 20.7‰ around 5100 cal a BP, which subsequently decreased by 1.1 to -19.6‰ around 1150 cal a BP, with a further 0.6‰ decrease during the last~900 years to a present-day average of 19.0‰. At~8200 cal a BP, Δ 13 C was 20.5‰, which is higher than at the present day, but lower than the peak at~5100 cal a BP. The Δ 13 C values of isolated fossil Cedrus pollen samples from the two different sites are closely aligned (Fig. 5C).

Aridity reconstruction
Summer aridity reconstruction (Fig. 5D) shows there has been a long-term overall trend towards drier conditions in the Middle Atlas region during the last~5000 years. The largest changes appear to occur around 5100 to~4000 cal a BP, and again from~1000 cal a BP to the present day. The severity of aridity appears higher when reconstructed using data from Dai (2011), although the pattern of aridity change is similar for both datasets. Present-day summer aridity is approximately 1 point lower (drier) on the scPDSI scale compared to conditions at~5100 cal a BP. At~8200 cal a BP, summer aridity is drier than at~5100 cal a BP, but it is more humid than the Late Holocene period. The reconstruction shows that summers in the Middle Atlas have always been in a state of drought, although the severity of this drought has increased.

Stable isotope composition of paired fossil material
The δ 13 C values from pollen concentrates and bulk sediment are notably different between Col du Zad and Lake Sidi Ali, while isolated Cedrus pollen from the two sites are remarkably similar. Modern Cedrus pollen collected near Col du Zad and Lake Sidi Ali present statistically similar carbon isotopic ratios (Bell et al., 2017). The closely aligned values in the fossil pollen data (Fig. 4 A) are consistent with a common prevailing climate at these   (Bell et al., 2017) and summer aridity (scPDSI), using normalized Δ 13 C values based on the method described in this study.  closely located sites. In contrast, bulk sediment contains a mix of organic material of terrestrial and aquatic origin, including different pollen types, spores, algae, plant tissue, insect fragments and charcoal, which may respond differently to the prevailing climate. Similarly, pollen concentrates include multiple pollen types as well as aquatic microfossils, which are difficult to separate out fully , thereby affecting the δ 13 C values. The greater inter-and intrasample variability in these sample types (Fig. 4B,C) probably results from the diversity of contributing species and mixing of terrestrial and aquatic carbon sources. We suggest that the strongly depleted isotopic values of pollen concentrates and bulk samples from Col du Zad reflect a relatively high contribution of organic matter from abundant aquatic macrophytes (Potamogeton, Ranunculus spp.), and the spring-fed meadow of terrestrial hygrophytes (Cyperaceae, Poaceae spp.), consistent with δ 13 C values for aquatic macrophytes without assimilation of bicarbonate (Keeley and Sandquist, 1992). In contrast, steep rocky slopes and lack of marginal plant communities (Zielhofer et al., 2017a) may underline the similar isotopic values of the three sample types at Lake Sidi Ali.
Overall, we consider that the isotopic signal of pollen concentrates and bulk sediment will reflect the assemblage  Table 2. Results of stable carbon isotope analysis, including: updated modern sporopollenin results for Lake Sidi Ali and Col du Zad (Bell et al., 2017), and fossil pollen (sporopollenin) results (this study). Corrected Δ 13 C results are used for the aridity reconstruction. (A) (B) (C) (D) Figure 5. Relationships between modern Cedrus atlantica sporopollenin (Bell et al., 2017) using normalized Δ 13 C values (this study) and the selfcalibrating Palmer drought severity index (scPDSI), based on: (A) CRU v3.24 scPDSI (Osborn et al., 2017) and (B) Dai (2011) datasets. (C) Results of stable carbon isotope analysis on fossil Cedrus pollen corrected for changes in CO 2 δ 13 C and CO 2 ppmv over time (Schubert and Jahren, 2012 composition and site conditions, rather than a robust climate signal, which is only attained by isolating a specific species and tissue type, as evidenced here by the isolated Cedrus pollen.

Reconstructed Holocene aridity from isolated fossil Cedrus pollen
Our findings from isolated fossil Cedrus pollen reveal that carbon isotope discrimination has decreased in the Middle Atlas region, and hence summer aridity has increased during the last 5000 years, indicating the climate has become drier. Around 5100 cal a BP, the climate was relatively humid compared to the present day, and similarly for around 8200 cal a BP, which has been described as a period of cool summers and dry winters . It is noteworthy also that fossil Δ 13 C values are outside the range of the site averages for the presentday Middle Atlas and are similar to trees growing in the Rif mountains today (Bell et al., 2017), further emphasizing a much more humid bioclimate than present in the Middle Atlas.
The widespread expansion of Cedrus forests in the Middle Atlas occurred before 5000 cal a BP (Fig. 6A), with maximum forest established between 5000 and 4400 BP at Lake Sidi Ali  and Col du Zad (Campbell, 2017). The gradual aridification of the area (Fig. 6C) appears to have promoted the long-term decline in forest cover during the late Holocene (4300 cal a BP to present). Conversely, the more humid summer conditions of the mid-Holocene and the period around 8200 cal a BP promoted the expansion of Cedrus atlantica in the Middle Atlas (Zielhofer et al., 2019).
Changes in forest density, here indicated by the arboreal/ non-arboreal pollen ratio (Magri, 1994), shows peak total forest occurred~300 years after aridity started to increase (Fig.  6B), while Cedrus forest continued to expand for another~300 years, stabilizing around 4300 BP before declining. This apparent lag may reflect the long-lived nature of the species (lifespan up to 1000 years) and its resilience to moisture stress, which is particularly true in older specimens (Linares et al., 2013). Importantly, the comparison may indicate the lagged population response to climate change as reflected in pollen assemblages versus a rapid response in isotopes. This lag effect may skew the interpretation of the timing of climate changes when based on pollen abundances alone. Our reconstruction falls within the reconstructed scPDSI values from tree-rings for the last~900 years in Morocco (Esper et al., 2007) and North Africa (Touchan et al., 2011), suggesting that the extremes of long-term change over the Holocene may have been similar in amplitude to the extremes of high-frequency (annual, decadal and centennial) variability. To date, we do not have sufficient sampling resolution for the last 1000 years to test for the wellknown dry to humid transition from the MCA to LIA as observed in tree-ring records (Esper et al., 2007).
A complexity for the study of past hydrological change in semi-arid regions is disentangling the influence of precipitation and temperature on drought regime. The summer aridity signal recorded in the carbon isotope composition of Cedrus pollen (Fig. 6C) may effectively integrate precipitation and temperature influence on summer soil moisture availability. Since C. atlantica is a shallow-rooted species (Aussenac, 1984), it lacks access to deep soil moisture resulting from winter precipitation. The choice of scPDSI here, and in the tree-ring studies as the reconstruction target, reflects the need to integrate temperature and precipitation impacts on bioclimatic moisture availability. However, as the aridification trend parallels the reduction in summer insolation (Fig. 6D), it would appear that increasing temperature was not the primary driver. If precipitation remained high throughout the Late Holocene, then temperature would not have been a limiting factor on leaf stomatal conductance, because soil moisture reserves would allow evaporative demands to be met through high transpiration. As such, the summer aridification trend could not simply be explained by rising temperature alone, and it is likely that decreasing precipitation is the main driver of the aridity trend detected here. This interpretation draws support both from the dominant influence of precipitation on C3 plant δ 13 C (Rao et al., 2017) and leaf physiological responses in trees in semiarid ecosystems (Grossiord et al., 2017).
Our results also show a similar trend to the ostracod δ 18 O record from Lake Sidi Ali (Zielhofer et al., 2017a) (Fig. 6E), and δ 13 C values of pollen concentrates from Lake Sidi Ali obtained during accelerator mass spectrometry 14 C dating   (Fig. 6F), particularly between~5000 and~2000 cal a BP. The main driver of change in δ 18 O over the Holocene is interpreted as variability in winter rainfall (Zielhofer et al., 2017a), suggesting a coherent annual climate signal (coupled summer and winter drying) during this period with an imprint on the pollen concentrate δ 13 C. Subsequently, the apparent divergence between δ 18 O and δ 13 C trends within the last 2000 years may suggest an increase in seasonal contrast, with increasing winter rainfall, but continuing summer drying. However, the mixed nature of material in the pollen concentrates makes them less representative of a seasonal signal, and thus far the resolution of the isolated Cedrus pollen is insufficient to confirm a change in seasonality. Overall, the comparison highlights the potential to decouple different seasonal climate signals by integrating multiple isotope proxies.

Practicalities and wider implications
Each pollen sample in this study provides a robust picture of the average regional climate conditions for the represented time period. By increasing the sampling resolution, there is great potential to enhance the understanding of bioclimatic changes not only for long-term trends but also of Holocene rapid climate changes. It is envisaged that for future analysis, the number of grains required for isotope analysis could be much smaller due to refinements to the technique and improvements in EA-IRMS sensitivities. Indeed, at 20 µg carbon, the number of fossil pollen required could be reduced by more than 1300 to~2700 grains, while at 5 µg carbon (the theoretical minimum) the total number of grains required is only around 650. The equivalent picking time would then be approximately 2-3 h, which is comparable to the time it takes to analyse a slide for traditional pollen analysis. An alternative approach to analysing pollen δ 13 C uses a spooling wire microcombustion device, which can analyse pollen samples using < 100 grains (Nelson et al., 2008;Nelson, 2012;Griener et al., 2013). However, this requires more specialized equipment, which is not as widely available as standard EA-IRMS setups.
More widely, our approach has exciting potential for new applications in other Mediterranean or global drought-stressed environments. This pollen-based isotopic aridity proxy can help to overcome uncertainties in the interpretation of vegetation records as to the drivers of change in pollen assemblagesa traditional challenge facing palaeoecological interpretation of the Late Holocene in the Mediterranean region (Jalut et al., 2009;Mercuri et al., 2011;Roberts et al., 2011). It may also help to disentangle the vegetation-climate disequilibrium (where vegetation composition may lag climate change) by comparison of the climatic signal from the stable isotope analysis, and the vegetation reconstruction from the pollen assemblagein the same samples and without chronological uncertainty. Furthermore, because pollen-based isotope reconstructions can be carried out over longer time periods than is possible with tree-ring studies, this technique can offer new insights into the dynamics of long-term climate changes. As yet, we do not know the prevailing summer aridity conditions during the Early Holocene, but we hypothesize dry conditions which favoured the expansion of deep-rooting sclerophyll trees . To date, our findings indicate a Mid-to Late Holocene aridification trend in northwest Africa, as found in the wider western Mediterranean region (Magny et al., 2002;Carrión et al., 2010;Roberts et al., 2011), and adjacent Sahara (Zielhofer et al., 2017b).

Conclusions
Our study offers new insights into climate change during the Late Holocene and documents the first use of stable carbon isotope analysis on fossil Cedrus pollen. Our study uses a new method to concentrate and isolate fossil Cedrus pollen from pollen assemblages that does not use traditional carbonbearing chemicals that would otherwise contaminate the isotope signal. Our findings indicate that summer aridity in the Middle Atlas has increased during the last 5000 years. The trend corresponds with an overall decline in arboreal pollen and Cedrus pollen, supporting a role for moisture stress as a key driver of regional Late Holocene vegetation change. Our findings also provide a tentative suggestion of centennial-scale disequilibria between total forest development and climate change.
The aridity trend corresponds to the decline in incoming summer solar insolation, and we suggest that the aridity trend was driven mainly by decreased summer precipitation, rather than increasing temperatures. Although thus far our results do not reveal short-term climate events as observed in the treering records, the overall range of reconstructed scPDSI values is comparable. Increasing the resolution of pollen-based isotope proxies would offer the same short-term insights provided by tree-ring records, but over longer periods. Overall, our findings help to confirm a summer aridification trend in north-west Africa and the wider western Mediterranean region since~5000 cal a BP and underline the value of further application of stable isotope research to fossil pollen.

Supporting information
Additional supporting information can be found in the online version of this article. Fossil Cedrus pollen isotope results and corrections applied. Modern Cedrus atlantica pollen normalized isotope results. Cedrus pollen data for Lake Sidi Ali and Col du Zad. Pollen concentrates and bulk sediment isotope results.